Freedom of speech and civic participation predicts national commitments towards animal protection

Abstract

Beyond our moral obligation to limit animal suffering, taking steps to improve animal welfare is essential for human wellbeing. Yet, the factors driving differences in animal treatment across countries remain unclear. Here, we used indicators of national policy and legislative commitments to animal protection, including enforcement mechanisms, alongside patterns of animal-product consumption and production to represent two dimensions of national commitments to animal treatment. We then examined how social, economic, and environmental factors are associated with variation across these dimensions. While high-income countries with lower dependence on agriculture may have more opportunity to focus on animal protection, our random forests analysis suggests that the level social freedom and civic participation among citizens (as captured by the Voice and Accountability indicator) can dramatically overwhelm this economic signal. In single-predictor regression models, Voice and Accountability also explained substantially more variation in animal protection than GDP per capita, accounting for 59% of the variation compared to 44%. Irrespective of economic conditions, structures that enhance political freedoms, civil liberties, and citizen participation in governance may provide necessary foundation for advancing legislation that protects animals.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-57055-7.

Introduction

Most mammals and birds exist within captive settings¹, while the remainder live in natural environments that are subject to growing human pressure². Directly or indirectly, human decisions influence the welfare of all animals on Earth. A growing body of evidence suggests that promoting animal welfare can enhance human and environmental wellbeing^3,4. By enhancing the living conditions of farm animals, for example, we could reduce global antimicrobial resistance^5,6, decrease risks of zoonotic and emerging diseases^7–11, and enhance animal product yield and quality¹². As such, understanding the factors that influence the quality of animal treatment across the globe is critical if we are going to design strategies to improve the wellbeing of both animals and humans.

A key theoretical framework for explaining cross-national variation in animal treatment is the concept of postmaterialism¹³. This perspective posits that, as societies achieve higher levels of economic security and satisfy basic material needs, individuals increasingly prioritize ‘postmaterial’ values such as environmental protection, social justice, and ethical concern for others¹³. Within this framework, concern for animal welfare can be understood as part of an expansion of moral consideration that accompanies postmaterialist value change¹⁴. As individuals become less preoccupied with basic needs, they may place greater emphasis on the ethical treatment of non-human animals, supporting stronger welfare standards, legislation, and consumption practices aligned with these values. Accordingly, economic development may influence animal welfare not only by increasing material capacity, but also by facilitating shifts in societal values that favour more humane treatment of animals.

Consistent with this perspective, previous research in animal welfare suggests that economic development and democracy may be important for establishing the financial and political foundations to improve animal treatment^15–18. With greater access to financial resources, high-income countries are expected to be in a better position to overcome any trade-offs between productivity and animal welfare¹⁹. In parallel, consumer attitudes towards animal welfare are thought to play an increasingly prominent role in regions where people have access to sufficient income^20,21 and education^21–23. As a result, activism related to animal welfare have grown in high-income countries, whereas in lower-income regions, other more proximal issues is still expected to take priority¹⁹. The economic structure of countries, particularly their dependence on agriculture may also play an important role in shaping these dynamics. In countries that are more dependent on agriculture, the financial risks associated with strict regulation are likely to discourage emphasis on animal welfare policies, whereas countries with lower agricultural dependence may have more opportunity to focus on food quality and animal welfare²⁴.

Beyond economic factors, cultural and social perspectives may provide a broader context underpinning national commitments to animal protection. Gender differences in attitudes toward animals are well documented, with women generally expressing more positive and compassionate attitude towards animals compared to men^25–27, suggesting that countries with lower gender inequality could have stronger animal protection policies. Similarly, greater access to equitable quality education may relate positively to national commitment towards better animal treatment, as opportunities for higher education are associated with greater concern for animal welfare²¹. Cultural and religious contexts can also contribute to cross-national variation, particularly through their influence on meat consumption patterns^28,29. However, the relative importance and interaction of these factors remain poorly understood, and the fundamental determinants driving differences in national commitments towards animal treatment remains unclear.

Here, we used indicators of national policy and legislative commitments to animal protection, along with patterns of animal-product consumption and production, to examine how socioeconomic and environmental factors explain variation in national commitments towards animal treatment. Our analysis draws on 13 indicators compiled from the ‘World Animal Protection International’ organisation³⁰, the ‘Global Animal law Association’³¹, and the ‘Animal Protection Institute’³². These indicators span a range of animal categories, including farm, companion, laboratory, wild, and other captive animals. While legal standards do not always reflect real-world practices, enforcement mechanisms are included in these assessments to help bridge this gap. Using multivariate approaches, we examined the main underlying dimensionality of national commitments towards animal treatment. We then use random forest models to assess the relative importance of socioeconomic factors in explaining variation across these dimensions. We also assessed the generalizability of our findings on an independent, earlier dataset covering 154 countries³³.

Results

Dimensionality of national commitments towards animal treatment

Based on our categorical principal component analysis, the scree plots revealed that only three dimensions are necessary to explain 79% of the variation in animal welfare commitments across these 50 countries. Notably, the first major axis was able to explain over 57% of the variation, broadly correlating with indices capturing regulatory frameworks related to the protection of both captive and wild animals (Fig. 1b-c). The strong alignment among these variables suggests that this first dimension can provide a standardized metric to reflect the overall level of animal protection, hereafter referred to as ‘Overall Animal Protection’. The second orthogonal axis of variation explained 12% of the variation across the dataset, and captured variation in per capita animal-product consumption and production, hereafter referred to as ‘Responsible Production and Consumption’ (Fig. 1b-c). While the third axis of dimensionality helped to explain a non-negligible proportion of the variation across countries, it did not strongly align with any specific indicators, as all loadings were below 0.6.

Fig. 1.

Main dimensions of national commitments towards overall animal welfare. (a) Spatial variation of the two main dimensions: Overall Animal Protection and Responsible Production and Consumption. (b) Loadings of each indicator on the main dimensions. We ordered the indicators in descending order according to their loadings on the first dimension. (c) Biplot of the categorical principal component analysis, where dots represent the raw data.

We projected the national data onto the subspace spanned by the two main dimensions to give two scores for each of the 50 countries allowing to visualise the spatial variation in the two major dimensions of national commitments towards animal treatment (Fig. 1a). No clear latitudinal trends emerged for the Overall Animal Protection dimension. The strictest levels of policy commitment were apparent in more developed countries in Europe, followed by New Zealand, and Mexico. However, the second axis – which relates to the per capita level of responsible consumption and production of farm animals – revealed opposing patterns to the first axis, with the highest values in Nigeria and India, followed by countries in Central and North Africa.

Socioeconomic drivers of the main dimensions

We used random forest models to explore how the two identified dimensions of national commitments towards animal treatment relate to national-level data on environmental, social and economic conditions. For the main dimension, reflecting national commitments towards animal protection (called Overall Animal Protection), there was almost no environmental signal, as socioeconomic factors ultimately determined the global pattern (see Fig. 2A). However, in contrast to common expectations^15,16,19,24, economic indicators including the GDP per capita and the ‘value added of the agriculture, forestry, and fishing to GDP’ only help explained a relatively small proportion of the variation in the Overall Animal Protection across the globe. Overall, high-income countries with lower economic reliance on agriculture tend to score higher in Overall Animal Protection (see middle and right panels in Fig. 2B). Interestingly, while high-income countries generally tended to have more stringent legislation, there was a large variation within the lower-income regions that could not be explained by the economic data (see shaded area of Fig. 3C). Indeed, there were many countries with relatively low per capita GDP, including Tanzania, India, Kenya, and Brazil, which scored more highly than wealthier counterparts in their commitments to Overall Animal Protection.

Fig. 2.

Variable importance and partial dependence plots of the three most important socioeconomic and environmental variables for the Overall Animal Protection dimension. (a) Normalized variable importance from the random forest model. Higher values indicate stronger reliance of the model on a given variable. Importance was estimated using permutation importance, defined as the decrease in model performance after randomly shuffling each predictor while keeping all other predictors unchanged. Model performance was measured using the mean squared error (MSE), and importances were normalized. Bars show mean ± SD across 10 random seeds. (b) Solid lines represent the partial dependence plots of the three most important variables, while black dots indicate the raw data; tick marks on the x-axis illustrate the distribution of each variable. (c) Two-way partial dependence plots illustrating potential interactions between the top variable and the next two most important variables, where darker colours would indicate higher Overall Animal Protection scores. To mitigate extrapolation risk during interpretation we overlaid the raw data on top of the plot.

Fig. 3.

(A,B) Relationship between Voice and Accountability or GDP per capita and Overall Animal Protection. Points represent observed values, lines show fitted values from single-predictor linear regression models, and shaded areas indicate 95% confidence intervals. (C) Overall Animal Protection dimension in relation to the GDP per capita with colours indicating levels of Voice and Accountability.

By far the strongest predictor of the Overall Animal Protection dimension was ‘Voice and Accountability’ (Fig. 2), which broadly reflects the freedom of speech and citizens influence on national policy. Countries with higher ‘Voice and Accountability’ showed the greatest tendency to prioritize animal protection, irrespective of economic status. This variable dramatically enhanced the performance of the random forests model, reducing the mean squared error by more than five times the reduction achieved by any other predictor.

To evaluate the robustness of the results across different statistical approaches, we also fitted a linear regression model including as predictors the three most important variables identified by the random forest. These predictors together explained a total of 60.8 % (adjusted R²: 57.96%) of the variance in the Overall Animal Protection dimension and findings were consistent with the random forest results, indicating that only ‘Voice and Accountability’ showed a significant effect (p < 0.001, t value = 3.95, =0.65, standard error = 0.16). For comparison, we also evaluated the variance explained by the two main predictors on their own, using single-predictor linear regression models, illustrated in Fig. 3. A-B. Voice and Accountability captured 58.6% of the variance (after adjustment for the number of predictors) on its own (Fig. 3A, p < 0.001, t value = 7.96, =0.77, standard error = 0.10), explaining patterns in the Overall Animal Protection dimension observed across both low- and high-income countries (Fig. 3C); whereas GDP per capita explained 44.44% of the variance (after adjustment for the number of predictors) (Fig. 3B, p < 0.001, t value = 6.02, =1.22, standard error = 0.20, log-transformed).

A similar analysis conducted on an independent older dataset covering 154 countries³³ yielded consistent results. Voice and Accountability and GDP per capita emerged as the strongest predictors, with the effect of the former substantially exceeding that of the latter (Fig. S3), providing support for the generalizability of our findings.

No predictors of comparable strength were identified for the Responsible Consumption and Production dimension, as the full model explained only 11.96% of the variance, compared to 56.02% for Overall Animal Protection dimension (Fig. 4A). Scores for this second axis generally correlated negatively with tertiary school enrolment, and positively with the economic dependence on agriculture (Fig. 4B). This pattern may mainly reflect differences in national development, rather than a direct effect of lower tertiary education leading to more responsible consumption and production. These results were not replicated on the larger dataset³³, where GDP per capita appeared as the strongest predictor (Fig. S3) although this discrepancy likely reflects differences in dataset composition, given that GDP per capita is incorporated into the construction of the index³³(but not into the Responsible Consumption and Production dimension).

Fig. 4.

Variable importance and partial dependence plots of the three most important variables for the Responsible Consumption and Production dimension. Similar to Fig. 2.

Discussion

By collating indicators of policy and legislation governing animal protection, along with patterns of animal-product consumption and production, we were able to examine the range of social, economic, and environmental variables that predict two aspects of national commitments to animal treatment. Our results does not support the common expectation that economic development determines the level of national commitments to animal protection^{15–17,19,24} and suggests that the extent of citizens influence on national policy and their freedom of speech can considerably overwhelm this economic signal. However, our results also suggest that legislative commitments to animal protection do not necessarily translate to improved responsibility of consumption and production, which generally tend to be highest in some lower-income countries.

Until now, the absence of large-scale, animal-based welfare data has precluded efforts to study factors influencing animal welfare across the globe. By relying on national policy and legislation and patterns of animal-product consumption and production, we were able to derive two main dimensions reflecting distinct aspects of national commitments towards animal welfare. One dimension captures minimum legal and policy standards applicable to animals, while the other captures the scale of animal use through production and consumption, reflecting the overall burden on animals. While these two dimensions do not directly measure animal welfare outcomes, they provide standardized insights into spatial variation in national commitment to animal treatment across the globe.

The first main axis was able to explain most of variation, broadly correlating with all indices relating to the protection of captive and wild animals, along with the strength of law against animal suffering. This axis – which we refer to as the Overall Animal Protection – tended to show the highest values in more developed countries. Specifically, the strictest levels of policy commitment to animal protection were in the more developed countries within the European Union, followed by New Zealand, and Mexico. In contrast, the second axis – which relates to the per capita level of responsible consumption and production of farm animals – revealed contrasting patterns to the first axis, with the highest values in Nigeria and India, followed by countries in Central and North Africa. These trends suggest that nations tend to show contrasting commitments to different aspects of animal treatment. While several of the more developed countries may have stringent legislation protecting animals, they do not appear to demonstrate the same level of commitment to responsible levels of per capita consumption of animal products, relative to lower-income countries with lower individual-level meat consumption rates³⁴.

By exploring the socio-economic factors correlating with our two main dimensions, we could explore the possible drivers of animal treatment across the globe. By far the strongest predictor of Overall Animal Protection dimension was ‘Voice and Accountability’, which broadly reflects the freedom of speech and citizens influence on national policy. This result contradicts the common expectation that economic development determines the level of national commitments to animal protection. Our results support the idea that higher income countries with lower dependence on agriculture tend to achieve higher scores in animal protection, as they may have a greater freedom to prioritise strict animal protection^15,16,19,24. However, our analysis suggests that the level of citizens’ social freedoms can considerably outweigh this economic signal. That is, the Overall Animal Protection dimension is highest in countries where citizens’ opinions are more likely to be reflected in national policy, irrespective of the economic status. This trend potentially suggests that the human condition may generally be favourable towards promoting good animal welfare, but that this perspective is only able to emerge in regions where citizens have a greater influence on national policy.

In contrast to the Overall Animal Protection dimension, we did not identify any strong predictors for the ‘responsible consumption and production’ dimension, as the model could only explain a marginal proportion of the variation across the globe. Altogether, these results potentially lends support to prior research, which suggests that changing the behaviour of individual consumers and producers may be more complex than shifting broader national-scale opinion and legislation^35,36.

The correlational nature of our analysis limits our capacity to discern causality in these relationships. Without time series data, it is difficult to establish causation. However, these correlations help uncover underlying factors that may shape global patterns of animal treatment and suggesting theories for future research to test. For example, it is possible that citizens in countries with high ‘Voice and Accountability’ have been responsible for driving strong commitments to protecting animals. However, it is also possible that the countries with higher concern for the welfare of animals are simply predisposed to promote societal freedoms of citizens. Future work could examine how national commitment in animal treatment change over time to explore causality and, ultimately, facilitate effective policy decision making.

While this study introduces a scalable framework for comparative analysis, it is constrained not only by the absence of longitudinal data but also by the resolution and coverage of available data. Future work should prioritize the development of higher-resolution, disaggregated datasets, expanding both geographic coverage (to include additional countries and subnational units) and granularity across different animal groups. Such refinements would enable more precise identification of drivers and heterogeneities in animal treatment practices.

In conclusion, the two identified dimensions provide unique standardized insights into national commitments towards animal treatment, capturing animal protection in the first dimension and responsible animal-product production and consumption in the second. Although developed countries tended to show the highest commitments to animal protection, they tend to lag in practicing responsible consumption and production. Yet, although the dimension related to animal protection showed moderate correlations with economic indicators, the tight coupling with ‘Voice and Accountability’ lends support to the idea that cultural and social perspectives may provide the broader context for national commitments to animal treatment. Irrespective of the economic context, social structures that enhance political freedoms, civil liberties, and citizen participation in governance may lay the necessary foundations for developing legislation to promote the protection of animals across the globe. Ultimately, this highlights potentially bright policy avenues for simultaneously improving animal and human wellbeing at national and international scales.

Methods

Dimensionality of national commitments towards animal treatment

We used 13 national indicators relevant to animal treatment derived from three data sources^31,32,37, covering 50 countries that includes major animal product-producing regions and corresponding predominantly to the year 2020. First, we used 10 animal protection indicators ranking the countries according to their legislation and policy commitments to protecting animals, published in 2020 by the ‘World Animal Protection International’ global organisation³⁷. The 10 indicators cover four main aspects: the recognition of animal sentience and prohibition of animal suffering (two indicators), the establishment of supportive government bodies (one indicator), and the support for international animal welfare standards (one indicator), and the presence of animal welfare legislation (six indicators: farm animals, companion animals, animals used for draught and recreation, laboratory animals, wild animals, and captive animals which includes zoo animals, private keeping of wild animals, and fur farming). The indicators scores countries from A (being the highest score) to G (being the weakest score). Secondly, we used the national database on animal legislations from the ‘Global Animal law Association’, based on the available online information as of 31st of March 2021³¹. Lastly, we used two indicators from the ‘voiceless animal cruelty index’ (VACI) developed by the ‘Animal Protection Institute’ in 2017 and updated in 2020³². These two indicators assessed countries based on the nature, extent, and severity of farmed animal cruelty while considering the quality of animal protection laws. Specifically, the ‘producing cruelty’ indicator, hereafter referred to as ‘responsible production’, scored countries based on the number of farmed animals slaughtered per capita, accounting for variations in animal treatment and protection. The ‘consuming cruelty’ indicator, hereafter referred to as ‘responsible consumption’, evaluates countries based on their consumption of farmed animals, examining the number of animals consumed per capita and the ratio of plant-based protein to farm-animal protein consumed.

Since we have both ordinal and continuous indicators, we performed a categorical principal components analysis³⁸ to identify the main dimensions explaining most of the variance. This method transforms the original ordinal variables into a numerical scale that maximize the variance explained by each component³⁹. Therefore, the number of dimensions affects the results, making it important to compare results across different dimensionalities when choosing the number of components. Thus, using the ‘gifi’ package and its ‘princals’ function⁴⁰, we examined scree plots for solutions ranging from two to six dimensions, suggesting three components. Before performing any analysis we discretized the Responsible Production and Consumption indicators by standardizing them, multiplying all values by 10, rounding, and adding a constant so that the lowest resulting integer value equals one³⁸. We then projected all observations onto the subspace spanned by the identified components to obtain one component score for each country per dimension.

Socioeconomic drivers of the main dimensions

To identify socioeconomic factors explaining variation in national commitments towards animal treatment, we selected a total of 15 variables spanning agriculture²⁴, economics^{15–17,19,24}, governance^17,18,41, education²¹, gender equality^25–27, and religion^28,29, as these domains have already been suggested to shape animal-related policies and practices. We also included environmental factors (average mean surface air temperature and precipitation from the Climate Change Knowledge Portal⁴² as they can influence political decisions and exert additional abiotic pressures on production¹⁶.

More specifically, we selected the gross domestic product (GDP) per capita and the annual percentage growth rate of GDP per capita to account for the level and recent trajectory of economic development. Agricultural variables comprised the percentage of land area that is used as agricultural land, and the value added of the agriculture, forestry, and fishing to GDP (in percentage). Governance was captured using three indicators: ‘Voice and Accountability’ capturing ‘perceptions of the extent to which a country’s citizens are able to participate in selecting their government, as well as freedom of expression, freedom of association, and a free media’, the ‘Rule of Law’ capturing ‘perceptions of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence’, and the ‘Government Effectiveness’ capturing ‘perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government’s commitment to such policie’⁴³. For education, we used the school gross enrolment ratio for primary, secondary, and tertiary education, defined as the ratio of total enrolment to the population of the relevant age group. For gender equality variable, we selected the ratio of female to male labour force participation rate (in percentage), where labour force participation rate refers to the proportion of the population ages older than 14 that is economically active. All of these variables were obtained from the world development indicators⁴². Finally, to represent religious composition, we included the proportion of Muslim and Hindus populations by country from the Pew Research Center⁴⁴, as these two religions have been specifically associated with differences in meat consumption patterns^28,29,45,46.

As first preprocessing step, we replaced missing values for the year 2020 for any data from the world development indicators with the mean values from 2019 to 2021. When these values were unavailable, we used linear interpolation based on the two most recent preceding years, excluding any data prior to 2017. We removed variables with more than 10% of missing data, namely, “school enrolment for the secondary”. To avoid multicollinearity, we iteratively removed the variable with the highest variance inflation score and recomputed the scores until all remaining variables were well below 10 (the highest value was 5.5), using the ‘Statsmodels’ python library⁴⁷. This process led to the removal of the ‘Rule of Law’ (VIF of 20.66) and ‘Government Effectiveness’ (VIF of 9.44) indicators (see Text S1 for additional analysis including these two variables). Lastly, we excluded any countries with at least one missing value, specifically, Myanmar, Egypt, Peru, and Venezuela. Data used for subsequent analysis included 46 countries and 12 variables. Because India represented an extreme outlier for the Hindu population variable (79.37%), whereas the median across included countries was only 0.04%, we report the results excluding India in the main text and provide the corresponding results including India in the Supplementary Information. These results were similar to the one reported in the main text (see Figs. S1–S2).

To contrast the impacts of these socioeconomic and environmental variables on the identified dimensions, we used random forests, as they are well suited for modelling non-linear relationships and handling datasets with a relatively large number of predictor⁴⁸. We used the Breiman’s Random Forest method⁴⁹ as implemented in the Scikit-learn python library⁵⁰. Model performance was assessed using leave-One-Out Cross-Validation a procedure well adapted to small dataset⁵¹ where each observation is considered once as the testing set, with the others as the training set. We evaluated model performance using the explained variance and used 10 different random seeds. Variable importance was assessed using permutation importance on models fitted to the full dataset using 10 different random seeds. Permutation importances were normalized within each seed to express relative importance, and the plotted values represent the mean ± SD across the 10 seeds.

To visualize the relationships between each variable and the dimensions we used partial dependence plot. These plots show the dependence between the identified dimensions and one or two variables of interest, marginalizing over the values of all other input variables. To evaluate the robustness of the results across different statistical approaches, we fitted a linear regression model with the main three variables identified by the random forest. We additionally fitted single-predictor linear regression models to assess the variance explained by each variable independently. Model assumptions, including normality and homoscedasticity of residuals, were verified, and variables were log-transformed where necessary to improve linearity. Since rule of law and government effectiveness were removed during the variable selection procedure due to high VIF, and both are correlated with our strongest predictor Voice and Accountability, we also examined their importance in explaining the overall animal welfare protection. We did this by considering each variable once individually and once in combination with the other predictors (see Supp. Fig. S3).

We assessed the generalizability of our findings using an independent dataset reflecting country conditions from approximately five years earlier (mostly around 2015, but underlying data spanning 2013–2018). We thus reran the analysis using socioeconomic factors from 2015 (see Supplementary Text S2). This dataset covered 154 countries and included two comparable variables related to animal treatment (called Political-commitment and Animal-use)³³. Although this dataset may not fully reflect current conditions, the underlying mechanisms and associated drivers are likely to have remained broadly consistent over time. A key limitation of this dataset is that both index dimensions, but particularly the second, are constructed from variables partly normalized by GDP, which may mechanically influence the relationship between GDP and the index, one of the central focuses of our analysis.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

T.W.C. provided overall project management, developed the data analysis and was involved in the writing of the initial draft manuscript. C.M. developed and performed the data analysis and was involved in the writing of the initial draft manuscript.

Data availability

All datasets used in this paper are publicly available, as referenced in the methods. The identified dimensions are provided in the supplementary information.

Code availability

The code used to perform the main analyses is available in the Supplementary information.

Declarations

Competing interests

The authors declare no competing interests.